한국어

Is the Learning Pyramid Accurate? What the Research Says

At a glance
  • The learning pyramid's numbers (5% from lectures, 90% from teaching others) did not come from any experiment.
  • Versions of the idea were in print by 1852, decades before anyone ran a memory experiment.
  • NTL Institute, one of the names most often attached to it, said it no longer has and cannot find the original research.
  • Active classes do beat lecturing, with 21.8% vs 33.8% failing in college science courses, but nothing like an 18-fold gap.
  • Whatever way you learn, closing the book and recalling is one of the best-supported ways to make it last.

The learning pyramid is not accurate. Its numbers, 5% retained from lectures, 10% from reading and 90% from teaching others, never came from an experiment, and an institute often credited with it says it cannot find the original research. The direction it points in is roughly right: doing beats only listening. But the gaps measured in real classes are modest, nothing like the 18-fold gap the pyramid implies (90% divided by 5%), and closing the book to recall is one of the best-supported ways to make learning last.

Where did the learning pyramid's numbers come from?

Nobody knows. NTL Institute, one of the names most often printed under the pyramid, has said it cannot find the research behind the numbers.

Letrud (2012) reprinted the reply NTL sent to people asking about the pyramid in 2009. NTL said the pyramid was developed and used at its Bethel, Maine campus in the early 1960s and that it believed the pyramid to be accurate, but that it no longer had, and could not find, the original research that supports the numbers. It added that many people had searched for that research and come up empty-handed. The numbers it summarized were: lecture 5%, reading 10%, audiovisual 20%, demonstration 30%, group discussion 50%, practice 75%, and teaching others or immediate use 90%.

Letrud and Hernes (2018) then searched digitized books and magazines for older versions. The oldest they confirmed was an 1852 line in a British magazine: what we read often fails to leave a lasting impression, what we hear finds no permanent place in memory, but what we see is engraved on the recollection. The first version with numbers appeared in 1906, and they found 63 versions published between 1906 and 1940.

Experimental research on memory began in the 1870s and 1880s (Ebbinghaus published his landmark study in 1885). The idea was in print decades earlier, so it cannot have come from that research.

The versions do not even agree with each other. Take listening alone:

VersionRetained from hearing (lecture)
Roads (1906), Sunday school magazine10%
Treichler (1967), audiovisual magazine20%
NTL (2009 reply)5%
Hansen (as quoted by Bravo et al. 2006)25%

Sources: Letrud & Hernes (2018), Letrud (2012). The authors suggest that the neat steps of 10% come from converting the "one tenth, three tenths" fractions common in the old versions. Claims you may see online, such as "retention after 24 hours" or "an MIT study", appear neither in Letrud and Hernes (2018) nor in NTL's reply as Letrud (2012) quotes it.

Is it Edgar Dale's cone of experience?

No. Dale's cone had no numbers, and it was not a ranking of how much people remember.

According to Letrud (2012), the educator Edgar Dale introduced the cone in his 1946 book as a way to sort learning experiences from concrete to abstract, and wrote that it was not intended as a hierarchy or rank order of learning processes (1946 edition, p. 47). None of the 1946, 1954 or 1969 editions contained percentages. At some point the old retention numbers were laid over a misreading of the cone, and that hybrid became today's pyramid. Masters (2020), reviewing medical education papers, put it bluntly in his abstract: the pyramid and its statistics are rubbish, and they do not come from Edgar Dale.

Why is it still everywhere?

Criticism has not slowed it down, at least in medical education, where some papers even cite the critiques as support for the pyramid.

Letrud and Hernes (2016) searched journals and encyclopedias published from 1990 to 2013. Some version of the pyramid appeared in at least 418 peer-reviewed articles and 11 encyclopedia articles, while 15 articles questioned or criticized it (counts as reported in their 2018 paper).

Far more papers printed the pyramid than questioned it

0100200300400500Publications · Printed it · 429429Printed itPublications · Questioned or criticized it · 1515Questionedor criticized it15
Number of publications, 1990 to 2013. "Printed it" is 418 peer-reviewed articles plus 11 encyclopedia articles, summed by us. The 15 that questioned it include one encyclopedia article.Source: Letrud & Hernes (2016), as reported in Letrud & Hernes (2018), section 1
See the numbers
Publications
Printed it429
Questioned or criticized it15

Medical education shows the same pattern. Masters (2013) found 43 medical education papers and conference papers from 1946 to 2012 that cited the pyramid. Repeating the same search five and a half years later (September 2012 to April 2018), he found 41 more (Masters 2020). New sources of the pyramid had appeared, refutations were being cited in its support, and even researchers who acknowledged its weakness still cited it.

Five and a half years after the first review, almost as many new citations

0102030405060Citing papers · 1946 to 2012 (about 66 years) · 43431946 to 2012(about 66 years)Citing papers · Sep 2012 to Apr 2018 (5.5 years) · 4141Sep 2012 to Apr 2018(5.5 years)41 in 5.5 years
Medical education papers and conference papers citing the pyramid. Both reviews searched the same databases (Academic Search Complete, CINAHL, Medline, Google Scholar) in the same way.Source: Masters (2013), Masters (2020), abstracts
See the numbers
Citing papers
1946 to 2012 (about 66 years)43
Sep 2012 to Apr 2018 (5.5 years)41

Letrud (2012) offers one guess at why it spreads so well. Its top and bottom rows neatly confirm something many educators already believe: that students should learn actively.

So is active learning a myth too?

No. The direction is roughly right. The size of the difference is not.

Freeman and colleagues (2014) pooled 225 studies of undergraduate science, engineering and math courses that compared traditional lecturing with active learning. Active learning meant students engaged through activities or discussion in class instead of only listening, for example group problem solving, in-class worksheets or clickers. Traditional lecturing meant continuous exposition by the teacher, with students taking notes.

Fewer students failed in active learning classes than in lectures

0%10%20%30%40%Failure rate · Lecture classes · 33.8%33.8%Lecture classesFailure rate · Active learning classes · 21.8%21.8%Active learning classes1.5 times morefailed in lectures
Average raw failure rate (%) in undergraduate science, engineering and math courses, across the 67 studies with failure data. Failing usually meant a D or F grade or withdrawing from the course. The 67 lecture groups alone had 29,300 students.Source: Freeman et al. (2014), Results and Fig. 1
See the numbers
Failure rate
Lecture classes33.8%
Active learning classes21.8%

Scores on exams and concept inventories were 0.47 standard deviations higher with active learning (158 studies). The authors translated that into about 6% on an average exam, enough to move a median grade from B− to B. That is a real difference. It measures something other than the pyramid's retention percentages, but it is nothing like the tenfold gap between a 5% lecture and a 50% discussion. The authors also note that the failure-rate studies lacked randomized designs, and the type and amount of active learning varied from course to course.

Older reviews could not find the pyramid's ranking either. As Letrud and Hernes (2018) describe, Henmon reviewed 29 memory studies from 1894 to 1910 in 1912 and could not even conclude that seeing beats hearing, while the kind of material kept making a difference. Lalley and Miller (2007), after reviewing the pyramid's methods, wrote that each led to retention and none was consistently better than the others.

Do you really remember 90% of what you teach?

Teaching does help you learn, and one experiment suggests much of that benefit may come from the recall you do while teaching.

Koh and colleagues (2018) split learners into four groups. One taught without notes, one taught from notes, one practiced recalling without teaching, and one solved arithmetic problems. On a comprehension test a week later, the group that taught without notes and the group that only recalled both did better than the other two. Teaching from notes did not give the same benefit.

So copying the top of the pyramid can miss the point. In that study, teaching from notes gave less benefit, and recalling with the book closed, with no audience at all, beat both teaching from notes and the control group. The Feynman technique note goes through this experiment in more detail.

What you can try today

Whether you listen, read or ask an AI, add three minutes of closed-book recall at the end. You only need a sheet of paper.

  1. When the lecture, article or AI answer is done, close the screen and the book.
  2. Write down three things you remember, as if explaining them to someone.
  3. Open it again and mark what you missed or got wrong. Tomorrow, turn only those parts into questions and recall them again.

Reading is not useless because a chart says 10%. It just leaves less behind when nothing follows it.

In Aloom

The point of this note is that what you do after learning matters more than the channel you learned through. In Aloom you ask an AI, keep only the answers worth keeping, and they connect to what you already know. Each day you recall 1, 3 or 5 of them as questions before seeing the answer, and each node's ink grows darker the more you recall it. Something you learned by reading lasts longer once it goes through that moment of recall. See how Aloom works

Frequently asked questions

Who created the learning pyramid?

No one knows. NTL Institute says it developed and used the pyramid in the early 1960s but cannot find the supporting research. Similar claims were in print by 1852, and numbered versions from 1906. Edgar Dale's cone of experience had no numbers.

Is the learning pyramid a myth?

Its numbers and ranking are. Studies do show that classes where students actively work tend to beat pure lecturing on average, but in undergraduate science courses the gap was about 6% on exam scores and failure rates of 33.8% vs 21.8%.

What actually works better than the learning pyramid suggests?

Recalling what you learned without looking (retrieval practice), and recalling it again after a gap, are well supported in learning research. Teaching helped most when people taught without notes and had to recall.

References